Cool, thanks for the info. I hadn't realized they had dipped their toes into the Crystalwell pool before Broadwell.
Haswell was the first one where they brought in the GT3e SKU, albeit only in a mobile incarnation. Broadwell was the first desktop chip to incorporate eDRAM, but Haswell was the first overall.
Indeed. Ivy Bridge to Haswell gave us ~5-7% IPC improvement. Haswell to Skylake gives us 5-20% IPC improvement, depending on the type of software being utilized.
Haswell was the first one where they brought in the GT3e SKU, albeit only in a mobile incarnation. Broadwell was the first desktop chip to incorporate eDRAM, but Haswell was the first overall.
It doesn't rule out the possibility of happening.
But its extremely unlikely. L2 cache is part of the core. They'd have to change the layout of the entire core to accomodate the doubled L2 cache, unless its physical array is 512KB in size and half is disabled already. Then they'd have to worry about keeping latency the same and power in control.
Even less likely because Servers and Mobiles have one thing in common: Power efficiency is a great boon to both. If they cut down associativity on client to lower power do you think they won't do that on server but rather turn around, keep the associativity the same and DOUBLE the size? You think they'd go with increased power, significant change in core layout and entirely seperate validation on the core for that?
Remember the expression "look at the forest not the trees"? Lowering power(like with reduced L2 associativity) to allow more multi-threading abilities on server, is just that. These kind of speculations just reduce the credibility of those saying them(referring to David Kanter).
Why do you think that doubled L2 would result in "significantly better" IPC? And what is "significantly better"? 5%? Because that's what "significant" means in CPU design. Likely it won't even net 3%, on modern architectures where caches have multiple tiers optimized for every code category, with L1 and L2 being very fast, but small ones. If the doubled L2 cache increases latency even by a little bit, then you'd lose most of the advantages of doubled capacity, because the point of L2 is fast access. L3 cache is a different story.
In fact, on servers if that increase in IPC means power use per core increases so much that you have to cut down on few cores, then they wouldn't use that IPC increasing feature.
AVX-512 is very likely in the client silicon. Just disabled. They are not going to build an entire new core, just to remove AVX-512.
Its no different than how its normally segmented, i3, Pentiums etc.
Do someone knows already if its possible to overclock via Base Clock Multiplier Locked Skylakes? What about non-Z Motherboards? A budget combo with a cheap Pentium and a mid range Motherboard (Low end Chipset with strong VRMs) could be a rather popular choice.
AVX-512 is very likely in the client silicon. Just disabled. They are not going to build an entire new core, just to remove AVX-512.
Its no different than how its normally segmented, i3, Pentiums etc.
That was one of the things mentioned in the video, AVX-512 is not in the client silicon at all, not just fused off.
Where have you been the last 3 generations? Check Sandy Bridge, Ivy Bridge and Haswell against Sandy Bridge-E, Ivy Bridge-E and Haswell-E Socket compatibility.Wait... if the client chip and the server chip are different... so that means that the server socket might be different than the client one?
Wait... if the client chip and the server chip are different... so that means that the server socket might be different than the client one?
Yeah, sadly they didn't left the chance to use dGPU on it...It wasn't mobile-only. The 4770R was a BGA desktop/SFF solution with CrystalWell eDRAM.
The Skylake Xeon is supposed to have 6 memory channels and over 3000 pins. That could raise the cost of the new Server and HEDT boards.
HEDT platform will use Socket R (LGA-2011) and only support 4 memory channels.
That is the current HEDT and server socket, the new sever socket is supposed to be Socket P. Where did you see it said that Skylake HEDT is no longer going to use the server CPU's and socket?
http://www.tomshardware.com/news/intel-skylake-compute-stick,30020.html
I didn't see these Skylake compute sticks mentioned.
If they can sell these for under $180-200, I'll buy a couple.
When i upgraded from i5 2500k 4.5Ghz to i7 6700K 4.5Ghz i was surprised how much FPS went UP.
I made complete test with 8 games and here is example:
Skyrim + mods
2500k 4.5ghz 38fps http://abload.de/img/tesv_2015_08_25_13_17tmsse.png
6700k 4.5Ghz 56fps http://abload.de/img/tesv_2015_08_25_19_150asse.png
mohit9206 said:So when can we expect Skylake i3 and Pentium reviews? I am aiming at either G4500,i3-6100 or an A8 from AMD.
My current cpu is Pentium G630 from 2011 so even the G4500 should be a good deal faster especially the HD530 graphics which i intend to use for gaming at 720p until i can save up for a decent graphics card.